It was very early when I wrote the image description, and the prompt wasn’t of much help. I did my best. lmao I realized couldn’t really figure out what was going on, but I was already halfway through writing the description.

It’s just about. It just needs a little nudge.

You scale the image down and then scale it back up with no interpolation, or use something like unfake.js to snap it to the grid. The author of this model also made a custom comfyUI node to interpret the image directly from the VAE and decode it directly into a pixel image.

Partial occlusions pretty consistently get weird like that.

This is definitely the type that grants wishes.